Microsoft and Meta are sharply reducing internal use of Anthropic Claude, even though Microsoft had projected annual Claude spending above $1 billion. Microsoft cut that spending by more than a third, while Meta halved Claude Code users from about 60,000 to 30,000. The shift matters because Anthropic is turning from a supplier into a direct rival for enterprise software budgets.
How the two companies reduced Claude usage
At Microsoft, executives Scott Guthrie and Jay Parikh instructed staff to favor internal tools such as GitHub Copilot and models from OpenAI. The monthly per-employee Claude budget in the cloud division fell from $100,000 to about $10,000, a tenfold reduction for that group. The directive links procurement policy to product strategy, steering demand toward systems where Microsoft controls distribution and margins.
At Meta, the decline came from two forces: workforce reductions and a deliberate push toward proprietary coding tools Muse Code and MetaCode. In a single 28-day period, Meta still spent more than $105 million on Claude Code alone, which shows how large consumption had become before the cuts. Meta had already introduced AI cost controls in June, so the latest move extends an existing savings program rather than starting one.
Cost pressure explains part of the decision, but competitive positioning explains the rest. Meta has limited incentive to fund Anthropic while preparing rival products, and it reportedly sought to restrict Anthropic access to its training data. Microsoft faces a similar conflict as Claude Cowork and ChatGPT Work develop into alternatives to Office workflows. Once a model vendor offers workplace suites, every internal license becomes support for a competitor.
What the split means for enterprise AI buyers
For companies using AI coding and office assistants, vendor overlap is becoming a selection risk. A provider that serves as a subcontractor today can release a competing interface tomorrow, changing pricing, data terms, and roadmap priorities. Smaller firms feel this through fewer neutral defaults, while large firms gain leverage to negotiate volume terms but must manage several internal standards at once.
The episode also changes how buyers should evaluate dependence on a single model. High consumption figures, such as nine-figure monthly outlays or six-figure user counts, make switching expensive and disruptive for engineering teams. What remains unclear is how much productivity changes when teams move from Claude to Copilot, Muse Code, or OpenAI models. Buyers should ask vendors about benchmark results, migration support, data-use boundaries, and whether internal usage commitments affect external pricing.
The marker to watch is whether Microsoft and Meta sustain lower Claude consumption while expanding their own tools across engineering and office work. Continued budget caps, wider deployment of Copilot, Muse Code, and MetaCode, and growth of competing work suites would confirm the break. That outcome would signal a market where hyperscalers both buy and displace leading models.
